By Grace Kim

The headline number in this story is structural, not a figure: Businesskorea reports that major Big Tech players are moving away from Nvidia — and frames that shift as a potential opening for Samsung Electronics and SK hynix.

That's a loaded claim for the memory market. Samsung and SK hynix supply the high-bandwidth memory (HBM — stacked DRAM that sits next to AI accelerators and feeds them data) that goes into Nvidia's top GPUs. Any change in how hyperscalers buy AI silicon ripples straight into Korean memory order books.

What does this change for Samsung and SK hynix?

Potentially a lot, potentially nothing yet. The report is a question, not a confirmed design win. If Google, Amazon, Microsoft or Meta accelerate in-house AI chips — the most common driver cited when Big Tech "moves away from Nvidia" — then HBM demand doesn't disappear. It shifts supplier mix. Samsung, which has fought for HBM qualification at Nvidia, could court hyperscaler internal programs instead. SK hynix, currently the HBM leader, would need to defend share on new sockets.

Why hyperscalers want out from under one vendor

• Nvidia controls the bulk of AI training accelerators, giving it pricing power hyperscalers resent.
• Custom silicon — Google's TPU line, Amazon's Trainium, Microsoft's Maia — cuts cost per token and locks in supply.
• Diversification protects AI roadmaps from single-vendor allocation crunches of the kind seen through 2023-2024.

Each of those custom chips needs memory. That's the Korean angle.

The caution flag

The source article, as surfaced, provides the thesis without disclosed specifications, order volumes or executive quotations. Treat the Samsung-SK hynix opportunity as market framing, not a booked contract. We'll follow with specifics — HBM3E (the current-generation high-bandwidth memory standard) shipment figures, qualification status, customer names — as they land.

For now, the trade is this: watch hyperscaler capex disclosures and any Korean memory guidance revisions. Those numbers will confirm or kill the thesis.